Mnemosis

把 AI 的记忆，从“无限存储 + 搜索”改造成“会记住、会遗忘、会整理、会自我怀疑”的系统。

Mnemosis is a human-inspired memory layer for AI agents. Most "AI memory" systems are just storage with semantic search bolted on: they save everything and recall by similarity. Mnemosis instead treats memory as a lifecycle — remembering, reinforcing, consolidating, forgetting, and reconciling — the way human memory actually works.

中文说明：README.zh-CN.md · English: README.md

Install

pip install git+https://github.com/liyexiaoyi/Mnemosis.git

Zero runtime dependencies (pure Python stdlib + SQLite). No server, no cloud embeddings required — optional embedder hooks only.

PyPI 版（pip install mnemosis）发布后会在这里同步更新。

SQLite 存储适合单进程/低并发场景；多个 Agent 并发写入时建议串行访问或接入 外部数据库适配层。

Related MCP server: aimemory

Quick start

from mnemosis import MemoryEngine
from mnemosis.types import MemoryKind, SourceRecord, SourceType

engine = MemoryEngine("memory.db")  # pass a path for persistence

engine.remember(
    "The user prefers Chinese for technical discussions.",
    kind=MemoryKind.SEMANTIC,
    source=SourceRecord(origin=SourceType.USER),
    cues=["user", "language", "preference"],
    importance=0.9,
)

for r in engine.recall("what language does the user prefer?", top_k=3):
    print(f"[{r.item.kind.value}] {r.score:.2f}  {r.item.content}")

check = engine.check("what is the user's favorite movie?")
print("knowledge gaps:", check.gaps or "none")

engine.sleep()  # offline consolidation: dedupe, promote, detect contradictions

中文快速开始

pip install git+https://github.com/liyexiaoyi/Mnemosis.git

from mnemosis import MemoryEngine
from mnemosis.types import MemoryKind, SourceRecord, SourceType

engine = MemoryEngine("memory.db")
user = SourceRecord(origin=SourceType.USER)

engine.remember(
    "用户喜欢用中文讨论技术问题。",
    kind=MemoryKind.SEMANTIC,
    source=user,
    cues=["语言", "偏好"],
    importance=0.9,
)
engine.remember(
    "昨天一起修了 SQLite 锁死的问题。",
    kind=MemoryKind.EPISODIC,
    source=user,
    cues=["SQLite", "锁死"],
)

for r in engine.recall("用户用什么语言聊天？", top_k=3):
    print(f"[{r.item.kind.value}] 相关度 {r.score:.2f}  {r.item.content}")

一分钟完整演示（记住 → 检索 → 新旧矛盾 → 睡眠整合 → 元认知 → 遗忘回收）：

pip install git+https://github.com/liyexiaoyi/Mnemosis.git
python examples/demo.py          # 仓库内

不想安装？直接在 Google Colab 打开演示笔记本。

Use with your AI client (MCP)

One-line MCP integration for Claude Desktop, Cursor, Codex and any MCP client:

{
  "mcpServers": {
    "mnemosis": {
      "command": "mnemosis-mcp",
      "args": ["--db", "/path/to/memory.db"]
    }
  }
}

Full guide (including Cursor and Codex configs): docs/mcp-quickstart.md.

Command line

mnemosis --db memory.db remember "用户喜欢用中文讨论技术问题。" --kind semantic
mnemosis --db memory.db recall "用户喜欢什么语言？"
mnemosis --db memory.db sleep
mnemosis --db memory.db check "用户最喜欢的电影是什么？"
mnemosis mcp --db memory.db   # or: mnemosis-mcp --db memory.db

Features

• Dual-track memory — episodic ("what happened") and semantic ("what is true") stored and recalled separately (complementary learning systems).

• Forgetting curve — memories decay with time; access and review strengthen them (Ebbinghaus + spaced repetition).

• Sleep consolidation — offline pass promotes repeated experiences into stable knowledge, prunes noise, dedupes, and detects contradictions.

• Source monitoring — every memory keeps origin, timestamp, trust, and confidence, so the system can say "I'm not sure" instead of confabulating.

• Active forgetting — unimportant memories fade; deletions go to a recyclable trash, never silently.

• Metacognition — confidence labels, contradiction reports, and knowledge gaps before the agent answers.

• Associative recall — memories are indexed by multiple cues (time, topic, people, keywords) and linked, so any angle can reach them.

• Pattern completion — a partial cue re-activates a whole integrated pattern (Rolls, 2013; Theves et al., 2024).

• Memory updating — update() revises facts, tracks revisions, and destabilizes before re-stabilizing (Nader et al., 2000).

• Chinese-optimized retrieval — CJK stopword filtering, zh date normalization, pinyin/English-mixed records, verified at 10k-memory scale (total hit@5 50.3% → 98.8%).

• Temporal reasoning — time-cell anchored ordering for "after/before/next" questions (Eichenbaum, 2014; Gilboa & Marlatte, 2017).

• Planning & reasoning memory — successful plans, steps, and outcomes are remembered and reused; failed steps are avoided on replan.

• Local-first, zero runtime dependencies — pure Python stdlib + SQLite.

Docs & research basis

• docs/memory-model.md — the memory model

• docs/research.md — human-memory papers behind each feature

• docs/architecture.md — architecture

• docs/mcp-quickstart.md — MCP integration

• docs/roadmap.md — roadmap

• CHANGELOG.md — every round of iteration, with measured results

Testing

python -m unittest discover -s tests -q   # 320 unit tests
python benchmarks/locomo_bench.py --mode keyword   # LoCoMo-style long dialogue

License

MIT. See LICENSE.

Contributing

PRs, issues and new benchmark scenarios are welcome — see CONTRIBUTING.md. Every change should come with a test or a measured benchmark result.